Poor nfs performance (in comarison to ftp)

b0fh · Guru Joined: 16 Jun 2003 Posts: 426

I've got two (infact more machines are in my network, but that doesn't matter) gentoo-machines running kernel 2.6 connected over 100mbit ethernet. Copying a large file from server to client over nfs gives me just about 3,5mbyte/s throughput. Both machines can give more, as a test over ftp (vsftpd) between these gave about 6,7mb/s throughput (file on same disk, about the same size).
I played around a bit with rsize/wsize options of nfs and tried nfsv2 and v3 - no better results.
So what might be the reason for this really bad nfs-performance? As nfs is a kernel-feature it shouldn't so terrible slow compared to ftp...

b0fh · Guru Joined: 16 Jun 2003 Posts: 426

I found a little tweak: Setting "RPCNFSDCOUNT" to 1 in /etc/conf.d/nfs pushes my transfer rate to about 5mb/s. Not as fast as I want, but better than before. But that also pushes the cpu-usage of the remaining process to about 60% (= 0%idle with disk-reads).
I also read here that other users have these performance-problems with gentoo (some say they didn't have the problem with oder distributions on same hardware). What else might be the reason?

b0fh · Guru Joined: 16 Jun 2003 Posts: 426

Still no ideas for this problem?

pilla · Posted: Tue May 25, 2004 3:25 pm Post subject:

How is the mount line that you are using in the clients? It should have rsize=8192,wsize=8192
_________________
"I'm just very selective about the reality I choose to accept." -- Calvin

b0fh · Guru Joined: 16 Jun 2003 Posts: 426

Yes, I tried these options (also different values for them), but it didn't increase my performance at all... What pushed it a little is the RPCNFSDCOUNT and setting "async" in /etc/exports...